A Mixed Hypergraph Convolutional Network for Session-Based Recommendation

Existing session-based recommendation (SBR) studies typically focus on capturing sequential dependencies using recurrent neural networks or modeling high-order relations through hypergraph convolutional networks. However, in real-world scenarios, sequential dependencies and high-order relationships simultaneously exist between items. To address this problem, we propose a novel SBR model named MHCN (Mixed Hypergraph Convolutional Network for SBR). MHCN proposed and constructed a new data structure—mixed hypergraph, which includes both directed hyperedges and undirected hyperedges, capturing the sequential relationships and higher-order dependencies among all items within a session, respectively. Then a mixed hypergraph convolutional network is designed to learn the representations of nodes in the mixed hypergraph. Finally, the soft attention mechanism is employed to obtain session representations. To validate the effectiveness of MHCN, we conducted experiments on two real-world datasets and the results demonstrate that integrating sequential dependencies and high order relations through a mixed hypergraph can effectively enhance SBR models.